The autoregressive model was used to establish the statistical model of daily average discharge, daily maximum tidal difference and daily average salinity. The daily characteristic value of saline tide was predicted. Gamma distribution function was introduced to characterize the memory effect of historical salinity in the model, and the continuous monotone attenuation trend was described by the importance coefficient. The prediction result is satisfactory, and the goodness of fit index of the measured and predicted daily mean salinity reaches 0. 8421. The gamma distribution function used in the model reflects the contribution of historical salinity and the time delay phenomenon of salinity, which has a good effect on improving the precision of salinity prediction.